Oh that Deco is clever, but a few things: 1) A diatribe <i>is</i> wordy so a 'wordy diatribe' is like saying muddy mud 2) I am not selling 3) No 'dogmatic positions', only explaining what he does not understand because of lack of experience, and what Deco doesn't understand for probably the same reason. 4) Arrogance does not suit a fool, and don't play word games when wrong. Just admit it.
Looks like you are in good (albeit fickle) hands here from the various contributors. But here is a further checklist of stuff that has not already been mentioned (apologies in advance if you are already well on top of these matters): a) Did you forward test (i.e. run the system against "live" real-time data, but without real executions)? If, yes, then for how long and how did this go? If not, why not? Forward test can give good forewarning that your strategy isn't based on a real edge, or has execution difficulties that mean it won't work "live" ... b) For your backtest ("probably between 100-200 instances of trades taken"), what were % winners, max consecutive losers, average trade (in % terms) and Sharpe Ratio? And what were these same numbers for your "cross validation" ("Cross validation used a sample of about 50-70 trades")? And for your "live" trading? c) How does your strategy do if you test it over more years of data (say 5)? In my view, this would be a good alternative check to testing against unseen data as you did ... d) Not wishing to appear insulting in any way, can I also check that you included the effect of commissions in all three stages?
I did not forward test before going live. The main reason is that since my backtesting is exactly like trading live as I had designed the system that way, there was no real point. Other than fills on entry that may lose a tick to slippage, there is no difference whatsoever between going forward and back. For my backtesting, I'm not going to post the detailed stats here. However, the cross validated test data did agree with the other set in terms of expected net profit and % of winners. In live trading, the % of winners is about the same but the net profit is smaller due to the winners being smaller than before. That being said, there was a 3 months period earlier that also had low performance, though it averaged to about 20% better profitwise. Unfortunately I did look into acquiring more data but is unable to do so (dont ask). Commissions have been accounted for during all comparisons.
There is a difference; 30-min bars have no internal structure, apart from Open-Hi-Lo-Close. Real data in a 30-min period does have structure; i.e. it is made up of ticks that go up, and down, and flat continuously throughout that 30-min period. Therefore your backtest engine will have made assumptions about the sequence of ticks in the 30-min bar, and therefore about how your orders got filled, and in many instances (I don't know how far apart your target and stop are from entry) this could make the difference between whether your target or your stop was hit first. Forward tests can help you get beyond this. Telling us what your % winners, max consecutive losers, average trade (in % terms) and Sharpe Ratio doesn't tell us anything about the specifics of your edge (I am sure others will correct me if this is untrue). If you go through other posts here on ET you'll see many, many instances where an OP discloses these or similar stats, without giving away their edge. However, if you are willing to list the stats I have suggested (and for backtest, and cross test, and "live"), I am quite sure you'll get lots of great insights from the comments made by the good folks here on ET. But that's your choice ... OK, I won't.
Please correct me if I'm wrong; however, your problem is not that you are losing with your strategy, but that it's not as profitable as you expect. Is that right? If so, that's o.k. You're on the right track. Test the Implied Volatility of your live trades with that of your backtested trades. Walt
Walt how do you test Implied Volatility? ALso I backtested manually using 5 minute bars when necessary. I told you the backtesting was exactly like trade.
I can't say that I'm 100 % systematic because I trade a portfolio of securities and they are selected, in part, using my discretion.
Sorry if you've posted this already, but are your stop and limit amounts the same amount (instead of an exit signal, that could result in different limit and stop amounts)? In forex, with fixed limit and stop amounts for every trade, I've found backtesting win ratios do not match forward trading because of what I call the "unwinnable case"; it might be called "whipsaw" but what do I know? In this case, with a larger limit amount than stop amount, what happens is from the entry point, the price goes in one direction by the stop amount or a little more; then, goes back the other direction by the stop amount or a little more. In this case, it doesn't matter if you went short or long; either direction would have been losers because the stop was hit in both directions. If I remove these cases, the win loss ratio goes to 50%, seems random but would be profitable with the larger limit amount. The frequency of unwinnable cases is approximated by limit and stop amounts as (limit - stop)/(limit + stop), but that formula is meaningless as there is a lot of variation from that estimate, and it doesn't hold up for all times of day. In backtesting the system gets caught on fewer of these cases than in forward testing. It means the system no longer applies, the market conditions changed, the system was overoptimized, the system was not a real edge to begin with, it can't figure out the difference between a trend and a range, or something else I can't think of to list here. Now if only I could find "unlosable cases".